摘要近年来，印刷体字符识别技术日益成熟，在现代生活得到了广泛使用，但是低质量印刷体字符识别技术的发展却仍存在很多难点。车牌图像字符是典型的印刷体字符，本文以车牌图像字符为研究对象，分析了低质量印刷体字符在识别过程中可能遇到的问题，设计了一个低质量印刷体字符的识别系统。84511

首先，本文对车牌图像字符进行预处理，讨论并研究了基于加权平均的灰度化算法、基于Roberts算子/Sobel算子的边缘检测等方法，重点介绍了基于边缘检测和数学形态学处理的车牌定位算法。其次，由于基本的字符分割算法对低质量印刷体字符的分割能力有限，采用阈值法对图像进行二值化，提出了一种改进的投影分割方法对字符进行分割，然后基于微结构法和像素法提取字符特征。最后通过基于BP神经网络的字符分类器，完成了印体字符的识别。该系统对车牌图像进行识别的实验结果表明，该系统的字符分割正确率可达93%，分类器识别的正确率可达90%。

毕业论文关键词：低质量印刷体；投影分割；BP神经网络；字符识别

Abstract In recent years, the technology of printed character recognition has become more and more mature。 The technology is widely used in modern life。 However, there are still many difficulties in the development of low quality printed character recognition technology。 The characters of license plate are typical printed characters。 In this paper, the image characters of license plate are chosen as the research object。 The problems that may be encountered in the process of recognition are analyzed。 A printed character recognition system is designed for the low quality printed character。

Firstly, the input image is pre-processed。 The methods used in this process such as the grayscale algorithm based on the weighted average and edge detection method based on Roberts operator / Sobel operator are discussed and studied。 Mainly introduced the license plate location algorithm based on edge detection and mathematical morphology。 Moreover, in view of the limited capability of basic character segmentation algorithm for low-quality printed character, threshold method is used for image binarization。 An improved projection segmentation method for the segmentation of low quality printed characters is proposed。 Then, the character feature is extracted based on micro-structure method and pixel method。 Finally, a character classifier based on BP neural network is designed in this paper。 The recognition for the printed character is completed by using the classifier。 The experimental results of this license plate image recognition system show that the segmentation accuracy rate of the system can reach 93% and the correct rate of the classifier can reach 90%。

Keywords: Low quality printed; Projection segmentation; BP Neural network; Character recognition 

目   录

第一章 绪论 1

1。1研究背景与研究意义 1

1。1。1研究背景 1

1。1。2 研究意义 2

1。2国内外研究现状 2

1。2。1 OCR技术的发展历史 2

1。2。2 OCR相关技术的研究现状 3

1。2。3 OCR技术的研究难点 4

1。3本论文的主要研究内容 5

1。4本论文的组织结构 6

第二章 图像的预处理技术 8

2。1原始图像预处理技术